Update app.py

app.py

@@ -1,26 +1,28 @@
 import os
 import gradio as gr
+import matplotlib.pyplot as plt
+from collections import Counter
 from google import genai
 from google.genai import types
 from google.genai.types import Part
 from tenacity import retry, stop_after_attempt, wait_random_exponential
 
-# Retrieve API key from environment variable
+# Retrieve API key from environment variables.
 GOOGLE_API_KEY = os.environ.get("GOOGLE_API_KEY")
 if not GOOGLE_API_KEY:
     raise ValueError("Please set the GOOGLE_API_KEY environment variable.")
 
-# Initialize the Gemini API client via AI Studio
+# Initialize the Gemini API client via AI Studio using the API key.
 client = genai.Client(api_key=GOOGLE_API_KEY)
 
-# Use the Gemini 2.0 Flash model as required
+# Use the Gemini 2.0 Flash model.
 MODEL_NAME = "gemini-2.0-flash-001"
 
 @retry(wait=wait_random_exponential(multiplier=1, max=60), stop=stop_after_attempt(3))
 def call_gemini(video_url: str, prompt: str) -> str:
     """
     Call the Gemini model with the provided video URL and prompt.
-    The video is referenced by its URI (expecting a publicly accessible URL) and passed as a Part.
+    The video URL is passed as a URI part with MIME type "video/webm".
     """
     response = client.models.generate_content(
         model=MODEL_NAME,
@@ -31,51 +33,88 @@ def call_gemini(video_url: str, prompt: str) -> str:
     )
     return response.text
 
-def analyze_video(video_url: str) -> str:
+def generate_chart(analysis_text: str) -> plt.Figure:
     """
-    Perform iterative, agentic video analysis.
-    In each iteration, the Gemini model refines its analysis based on previous output.
+    Create a simple bar chart based on the frequency of selected keywords in the analysis.
+    """
+    # Define keywords of interest
+    keywords = ["suspicious", "anomaly", "incident", "alert", "object", "movement"]
+    # Lowercase the analysis text and split into words
+    words = analysis_text.lower().split()
+    # Count occurrences for each keyword
+    counter = Counter({kw: words.count(kw) for kw in keywords})
+    
+    # Create a bar chart using matplotlib
+    fig, ax = plt.subplots(figsize=(6, 4))
+    ax.bar(counter.keys(), counter.values(), color="skyblue")
+    ax.set_title("Keyword Frequency in Analysis")
+    ax.set_ylabel("Count")
+    ax.set_xlabel("Keyword")
+    plt.tight_layout()
+    return fig
+
+def analyze_video(video_url: str, user_query: str) -> (str, plt.Figure):
+    """
+    Perform iterative (agentic) video analysis.
+    The analysis is refined over several iterations, incorporating the user query if provided.
+    Returns a Markdown report and a matplotlib chart.
     """
     analysis = ""
     num_iterations = 3
 
     for i in range(num_iterations):
+        base_prompt = "You are a video analysis agent focusing on security and surveillance. Provide a detailed summary of the video, highlighting key events, suspicious activities, or anomalies."
+        if user_query:
+            base_prompt += f" Also, focus on the following query: {user_query}"
+            
         if i == 0:
-            prompt = (
-                "You are a video analysis agent focusing on security and surveillance. "
-                "Provide a detailed summary of the video, highlighting any key events, suspicious activities, or anomalies."
-            )
+            prompt = base_prompt
         else:
-            prompt = (
-                f"Based on the previous analysis: \"{analysis}\". "
-                "Provide further elaboration and refined insights, focusing on potential security threats, anomalous events, "
-                "and any details that may help a security team understand the situation better."
-            )
+            prompt = (f"Based on the previous analysis: \"{analysis}\". "
+                      "Provide further elaboration and refined insights, focusing on potential security threats, anomalous events, "
+                      "and details that would help a security team understand the situation better. ")
+            if user_query:
+                prompt += f"Remember to focus on: {user_query}"
+                
         try:
             analysis = call_gemini(video_url, prompt)
         except Exception as e:
             analysis += f"\n[Error during iteration {i+1}: {e}]"
-            break  # Exit if an error occurs
-    return analysis
+            break
+
+    # Create a Markdown report (adding headings and bullet points if desired)
+    markdown_report = f"## Video Analysis Report\n\n**Summary:**\n\n{analysis}\n"
+    
+    # Generate a chart visualization based on the analysis text.
+    chart_fig = generate_chart(analysis)
+    return markdown_report, chart_fig
 
-def gradio_interface(video_url: str) -> str:
+def gradio_interface(video_url: str, user_query: str) -> (str, any):
     """
-    Gradio interface function that takes a video URL and returns the analysis.
+    Gradio interface function that takes a video URL and an optional query,
+    then returns a Markdown report and a visualization chart.
     """
     if not video_url:
-        return "Please provide a valid video URL."
-    return analyze_video(video_url)
+        return "Please provide a valid video URL.", None
+    return analyze_video(video_url, user_query)
 
-# Define and launch the Gradio interface
+# Define the Gradio interface with two inputs and two outputs.
 iface = gr.Interface(
     fn=gradio_interface,
-    inputs=gr.Textbox(label="Video URL (publicly accessible, e.g., YouTube link)"),
-    outputs=gr.Textbox(label="Security & Surveillance Analysis"),
+    inputs=[
+        gr.Textbox(label="Video URL (publicly accessible, e.g., YouTube link)"),
+        gr.Textbox(label="Analysis Query (optional): guide the focus of the analysis", placeholder="e.g., focus on unusual movements near the entrance")
+    ],
+    outputs=[
+        gr.Markdown(label="Security & Surveillance Analysis Report"),
+        gr.Plot(label="Visualization: Keyword Frequency")
+    ],
     title="AI Video Analysis and Summariser Agent",
     description=(
         "This agentic video analysis tool uses Google's Gemini 2.0 Flash model via AI Studio "
-        "to iteratively analyze a video for security and surveillance insights. It makes repeated "
-        "LLM calls to refine its analysis of the video content."
+        "to iteratively analyze a video for security and surveillance insights. Provide a video URL and, optionally, "
+        "a query to guide the analysis. The tool returns a detailed Markdown report along with a bar chart visualization "
+        "of keyword frequency."
     )
 )